Read the full analysis: Why The AI Race Is Driving A New Data Center Power Strategy on ThorstenMeyerAI.com
TL;DR
Two October developments illustrate how the AI buildout is changing data center power planning: Vantage reserved more than 1 gigawatt of factory capacity for Heron Power’s solid-state transformer, and NiSource agreed to a 20-year power deal with Microsoft that includes new generation and a 600-megawatt battery. The deals point to interest in higher-voltage direct-current systems and storage, but they do not establish how widely these designs will be adopted or how their costs will affect electricity customers.
Recommended by digitechbytes.com · AI-assisted guides
Shopping for technology news & gadgets? Start with the guides we keep up to date:
Updated October 202615 Best Portable Power Banks for Phones in 2027See the top picks →
Data center power planning is shifting as AI workloads require denser racks and more electricity: Vantage reserved more than 1 gigawatt of factory capacity for Heron Power’s solid-state transformer, while NiSource signed a 20-year agreement with Microsoft that adds about 1,100 megawatts of generation and a 600-megawatt battery. The deals show companies addressing power supply and distribution alongside computing hardware, though the full costs and wider effects on electricity bills remain unsettled.
Vantage’s reservation, reported on October 21, concerns factory capacity for Heron Power’s transformer, which the source describes as converting grid electricity directly to 800-volt direct current. A reservation of manufacturing capacity is not the same as a confirmed installation at a particular data center. Heron’s design also incorporates storage to help buffer changes in power demand, according to the source.
On October 16, Indiana utility NiSource signed a 20-year power agreement with Microsoft. The deal is reported to add about 1,100 megawatts of generation and include a 600-megawatt battery. The source does not specify the project’s full construction schedule, cost allocation or how the power will be shared with other customers.
The engineering rationale is that conventional data center systems convert power several times before it reaches servers, with energy losses producing heat. NVIDIA’s 800-volt reference design proposes converting grid power to DC near the facility and distributing it to high-demand racks. NVIDIA says its approach could use 45% less copper than 415-volt AC distribution and deliver up to 5% better end-to-end efficiency than current 54-volt systems. Those figures are vendor claims, not independent measurements.
Power architecture
Infographic 01 · From grid to rack
AI racks near a megawatt are outgrowing the AC power chain.
Why the industry is moving to 800-volt direct current: fewer conversions, less copper, batteries on the main path. The numbers are vendor figures with different baselines, so they cannot be added up.
Today: many conversions
Typical AC chain from the grid to the rack
Medium-voltage AC from the gridTransformer steps it down to low-voltage ACUPS converts to DC for the batteries and back to ACDistribution units feed the racksEvery server power supply converts AC to DC again
Each stage loses energy, and almost all of the loss becomes heat the cooling system must remove.
800 V DC: convert once, outside
NVIDIA reference design, full-scale production planned with Kyber racks in 2027
13.8 kV AC from the gridConverted once to 800 V DC at the edge of the site, batteries on the same DC bus800 V DC busway to the racksStepped down inside the rack, for 1 MW racks and beyond
Bridge for existing halls: a “sidecar” cabinet next to the rack converts AC to 800 V DC, while the old chain stays in place.
Power lost per MW delivered to the rack
kW, SMA white-paper model vendor model
AC chain, conversion at the racktraditional design~80 kW800 V DC from outside the buildingmedium voltage converted once~20 kW
What NVIDIA says 800 V DC changes
NVIDIA’s engineering claims claim
Sources: NVIDIA technical blog (20 May 2025, updated 31 Jul 2025); SMA America, sponsored content in Utility Dive (26 Oct 2026). Rack density: IEA, Key Questions on Energy and AI (Apr 2026) (AI server power density 11× from 2020 to 2025, a further 4× by 2027).
Powered by Thorsten Meyer AI
Power Limits Shape AI Expansion
For operators, the constraint on expanding AI capacity is increasingly not just whether they can obtain chips, but whether they can secure electricity, grid connections and electrical equipment. The source cites International Energy Agency estimates that data center electricity use worldwide could rise from 485 terawatt-hours in 2025 to 950 terawatt-hours in 2030. That projection signals the scale of expected demand; it does not mean every region will experience the same growth or pressure.
Higher-voltage DC distribution could reduce some conversion steps and the amount of copper required inside facilities, while batteries could help manage short-term load swings. But savings and performance depend on site design and operating conditions. The NiSource-Microsoft agreement also highlights a separate issue: new generation and storage can support a large customer, while the division of project costs and benefits among the company, utility and other ratepayers depends on the contract and regulatory arrangements.
Grid access can take longer than building computing equipment. The source cites estimates that transformers may take two to three years to arrive and grid connections may take five to ten years in some places. Those delays can shape where data centers are built and when their capacity becomes available.
From AC Systems to Dense AI Racks
Data centers have traditionally received alternating-current power, stepped it down through transformers and other equipment, then converted it to low-voltage DC near servers. This architecture served racks drawing tens of kilowatts, but AI systems are pushing rack demand much higher. The International Energy Agency says AI server power density increased 11-fold between 2020 and 2025 and could quadruple again by 2027. It estimates a rack could then draw peak power comparable to 65 households.
NVIDIA’s proposed design converts 13.8-kilovolt grid power to 800-volt DC at the facility perimeter, with the company planning full-scale production alongside its Kyber systems in 2027. A separate Utility Dive article published October 26 was sponsored by inverter maker SMA America. It describes a “sidecar” conversion cabinet that could let 800-volt racks operate in existing AC data halls, but the article’s modeled efficiency figures are not independent field results.
Storage is part of the same shift because batteries deliver DC power and can respond to rapid changes in demand. The IEA estimates data centers could have 20 to 25 gigawatts of batteries by 2030, and says storage could potentially serve as a grid asset if incentives support that role. This is a forecast, not a confirmed deployment total.
“A 1 MW rack at 54 volts could require up to 200 kg of copper busbar.”
— NVIDIA, in its 800 VDC technical blog
Costs, Deployment and Grid Effects
The reported deals do not establish how many data centers will adopt 800-volt DC systems, when Heron’s reserved factory capacity will translate into installed equipment, or how the design will perform at scale. NVIDIA’s copper, efficiency and maintenance figures are company estimates, while the SMA-sponsored article presents a vendor’s modeled case rather than independent operating data.
The NiSource agreement’s reported capacity figures do not resolve how its generation and battery will be financed, what electricity prices Microsoft will pay, or whether costs could be passed to other customers. National and regional electricity statistics also do not show that data centers alone caused any particular increase in household bills. The source reports average U.S. residential electricity prices of 18.21 cents per kilowatt-hour in 2026, compared with 17.30 cents in 2025, but does not establish the share of that change attributable to data center demand.
It is also unclear how quickly utilities can deliver new connections. A large interconnection queue does not mean every proposed project will be built: the source cites Berkeley Lab data showing about 2,300 gigawatts of U.S. requests at the end of 2024, while only 13% of earlier requests were completed.
Projects Test the New Power Model
The next practical test is whether the announced supply and equipment plans reach construction and operation on the schedules their developers set. Vantage’s capacity reservation will need to be followed by details on production, orders and deployments; the NiSource-Microsoft agreement will need further project information on generation, battery development and timing.
For the broader shift, operators and utilities will be watching the move toward 800-volt racks, including NVIDIA’s stated 2027 production plan, and whether sidecar systems can fit into existing facilities. Independent performance data, utility approvals and clearer cost arrangements will help show whether these designs ease power constraints or shift costs elsewhere. Until then, grid access and equipment lead times remain major variables in how quickly AI data center capacity can grow.
Key Questions
What did Vantage reserve?
Vantage reserved more than 1 gigawatt of factory capacity for Heron Power’s solid-state transformer, according to the source. That reservation does not by itself confirm installations at specific data centers.
What is 800-volt DC power meant to change?
The design converts power to 800-volt direct current nearer the facility entrance and distributes it to high-demand racks. Its proponents say this can reduce conversion losses and copper use, but cited performance figures are vendor claims.
What is included in the NiSource-Microsoft agreement?
The reported 20-year deal adds about 1,100 megawatts of generation and includes a 600-megawatt battery. The source does not provide full project costs, construction dates or details on how expenses may affect other customers.
Will data center batteries lower household electricity bills?
That is not established. Batteries may help manage data center demand and could support the grid under suitable incentives, but the effect on household prices depends on project costs, utility rules and how the storage is operated.
Source: ThorstenMeyerAI.com